This paper describes a new method to select a non-uniform set of sample points specifying a color transformation in a more accurate way than achievable through the same number of uniformly distributed sample points. Furthermore, a specific accuracy can be achieved with far fewer non-uniform sample points. This has direct implications on the real-time processing performance and on the resources (ink, media, time) needed if the set of sample points induces a customized test chart for printer characterization. Although many practitioners implement color transformations as a piecewise linear interpolation using an n-dimensional Delaunay tessellation, this paper discusses the potential to achieve better results with a data-dependent tessellation.